Birth weight and body fat mass in adults assessed by bioimpedance in the ELSA-Brasil study

Intrauterine life is a critical period for the development of body fat and metabolic risk. This study investigated associations between birth weight and total and truncal body fat in adults. To do so, we analyzed data on 10,011 adults participating in the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil) who self-reported birth weight as < 2.5kg, 2.5-4.0kg, or > 4.0kg at baseline (2008-2010) and underwent bioimpedance in the next follow-up visit (2012-2014). Greater mean total and truncal fat mass were seen in those with high birth weight compared with adequate birth weight (p < 0.001) in both sexes (total fat: 25.2 vs. 23.1kg in men and 31.4 vs. 27.7kg in women, and truncal fat: 13.5 vs. 12.4kg in men and 15.9 vs. 14.2kg in women). U-shaped patterns were observed in restricted cubic-spline analyses in the subset of 5,212 individuals reporting exact birth weights, although statistically significant only for those with high birth weight. In the whole sample, in comparing high to adequate birth weight, the latter predicted having a large (> 85 percentile) total and truncal fat mass, respectively: OR = 1.76, 95%CI: 1.37-2.25 (men) and OR = 1.86, 95%CI: 1.42-2.44 (women); OR = 1.68, 95% CI: 1.31-2.16 (men) and OR = 1.73, 95%CI: 1.31-2.28 (women). However, low birth weight predicted having a large (> 85 percentile) % truncal fat only in women (OR = 1.40, 95%CI: 1.03-1.91). In conclusion, in these men and women born in a period in which fetal malnutrition was prevalent, birth weight showed complex, frequently non-linear associations with adult body fat, highlighting the need for interventions to prevent low and high birth weight during pregnancy. Obesity; Birth Weight; Body Composition; Adipose Tissue; Observational Study Correspondence V. C. Luft Programa de Pós-graduação em Epidemiologia, Faculdade de Medicina, Universidade Federal do Rio Grande do Sul. Rua Ramiro Barcelos 2400, Porto Alegre, RS 90035-002, Brasil. vcluft@hcpa.edu.br 1 Faculdade de Medicina, Universidade Federal do Rio Grande do Sul, Porto Alegre, Brasil. 2 Hospital das Clínicas de Porto Alegre, Porto Alegre, Brasil. 3 Instituto Oswaldo Cruz, Fundação Oswaldo Cruz, Rio de Janeiro, Brasil. 4 Centro de Ciências da Saúde, Universidade Federal do Espírito Santo, Vitória, Brasil. 5 Faculdade de Medicina, Universidade Federal de Minas Gerais, Belo Horizonte, Brasil. This article is published in Open Access under the Creative Commons Attribution license, which allows use, distribution, and reproduction in any medium, without restrictions, as long as the original work is correctly cited.


Introduction
Obesity, a major public health problem, is now the world's 6th most important risk factor for overall disease burden 1 . It presents a high and growing prevalence in Brazil and the world 2 , and causes hyperglycemia, hypertension, hypercholesterolemia, sleep apnea, and leads to type 2 diabetes, cardiovascular diseases, many cancers, chronic renal disease and a series of other conditions, including musculoskeletal ones, producing disease burden 1,3,4 .
Obesity results from a complex multifactorial process, which involves not only lifestyle but also environmental exposures, social factors, socioeconomic conditions, family history and genetic factors 5 . Weight gain is manifested throughout the course of life, and obesity risk begins early in life, with the intrauterine period being critical in its development 6,7,8 . Both low and high birth weight have been linked to greater risk of obesity in adulthood 9 , as well as to obesity-related metabolic problems such as diabetes and hypertension 10 . There is compelling evidence that intrauterine adversity not only restricts fetal growth, but also alters gene expression in a manner which favors increased future risk of adiposity and metabolic diseases 11,12 .
Body fat distribution, especially central obesity, has been shown to predict mortality better than overall adiposity 13 . Some studies have shown a relationship between birth weight and the distribution of body fat in adult life 14,15 , but understanding on the development of different aspects of obesity within a life course perspective is still incipient. The distribution of body fat logically differs between men and women in response to the specific needs and stressors of males and females throughout evolution 16,17,18,19,20 . Studies in human populations of low or high birth weight with body fat distribution in adult life found that responses to adverse exposures in early life are gender-specific 21,22,23,24,25,26,27,28,29 . Previous analyses of the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil) have shown gender-specific associations between birth weight and greater overall and central obesity, assessed by body mass index, waist circumference and waist-to-hip ratio 30 and additionally diabetes 31 in adult life. Almost all of the cited studies, however, have been based on anthropometric measurements. We are unaware of prior studies investigating how birth weight relates to adult adiposity obtained with bioimpedance.
Although anthropometric measurements may provide a useful indication of central obesity in a clinical setting, most studies have concluded the estimation of visceral fat using anthropometry is limited, accounting for between 50% and 80% of the variance in both sexes 32 . Emerging evidence points to the existence of many "obesity phenotypes", with differing associations to chronic diseases 33 . Bioimpedance analysis may contribute to a better understanding of adiposity distribution. In this context, this study aims to investigate the association of birth weight with total and truncal body fat, as evaluated by bioimpedance in adults.

Method Subjects
This study reports a cross-sectional evaluation from the ELSA-Brasil, a cohort study whose main objective is to investigate factors related to the etiology of diabetes and cardiovascular diseases. Participants are active or retired civil servants from universities or research institutions, recruited between the ages of 35 and 74 years in capital cities of six Brazilian states: Rio de Janeiro, São Paulo, Rio Grande do Sul, Minas Gerais, Bahia, and Espírito Santo 34 . Volunteers were invited through announcements and telephone calls, being the latter more commonly used to attract retirees. A randomly ordered list of employees, stratified by sex, age, and occupational category, was also used to actively recruit eligible participants. Efforts were made to recruit similar proportions of men and women, as well as predefined proportions of age groups and occupational categories 34 .
Of the initial sample of 15,105 participants at the cohort's baseline (2008)(2009)(2010), 223 died between visits and 868 did not attend the follow-up visit (2012-2014). Additionally, 2,006 were excluded because of uncertain birth weight, 189 for twin birth, 650 for prematurity or uncertainty about pre-Cad. Saúde Pública 2021; 37(2):e00061619 maturity, 741 for lack of valid bioimpedance data, 122 for previous bariatric surgery, and 295 for lack of information on covariates, leaving a total sample of 10,011 participants for analysis.
During the baseline exam, we asked participants to indicate their birth weight as "less than 2.5kg" (low birth weight), "2.5-4.0kg" (normal birth weight), "greater than 4.0kg" (high birth weight) or "I do not know", in addition to their exact birth weight, if known.
At the follow-up visit, in the study center, body composition was evaluated by segmental bioimpedance with 8 electrodes (InBody 230; InBodyCo., Seoul, Korea), capable of distinguishing trunk and limb masses. Participants were instructed to fast for at least 8 hours, and not to ingest caffeine nor engage in physical activity within the 24 hours prior to examination. The participants urinated within minutes prior to measurement, wore standardized clothing and removed all metallic objects such as earrings, rings, glasses, etc. Bioimpedance was not performed on participants with pacemakers or metal prostheses.
Four outcome variables were defined based on fat mass as determined by bioimpedance: total body and truncal fat mass (kg), and the relative proportions (%) each comprised of total body weight.
Information concerning relevant covariates -preterm and twin birth, maternal and paternal history of diabetes, educational level, net family income per capita and age 34 -were obtained through interviews. Additionally, height, weight, waist and hip circumferences, and sagittal abdominal height were concomitantly measured. The staff obtaining these data were previously trained and certified, following ELSA-Brasil's quality assurance and control routines 35 .

Statistical analysis
We describe the characteristics of participants as absolute and relative frequencies for categorical variables, and mean (standard deviation) or median (25-75 percentile) for continuous ones. Scatter plots and Spearman correlation coefficients were used to describe the relationship between body composition when estimated by bioimpedance and anthropometry (body mass index -BMI, waist circumference, hip circumference, waist-hip ratio, waist/height, sagittal abdominal height).
The distribution of covariates across birth weight categories, testing crude differences with the chi-square test for categorical variables and analysis of variance for continuous ones, is presented separately for men and women. Means of total fat mass (kg), total fat mass/weight (% body fat), trunk fat mass (kg) and trunk fat mass/weight (% truncal fat), are presented by sex, according to birth weight and to categories of age, net family income per capita, race/color, educational level, mother's educational level, and maternal and paternal history of diabetes, adjusted in analysis of variance.
High values of body composition were defined as those above the 85th percentile of the distribution of each body composition variable and estimated the odds of those reporting low or high birth weight, compared to appropriate birth weight, to present these high values, adjusting through logistic regression for covariates.
In futher analyses, in participants who informed their specific birth weight (n = 5,212), we employed restricted cubic spline multiple linear regression, with nodes defined at percentiles 10, 50 and 90 36 to estimate the differences in total or truncal body fat, as a percentage of total weight, across the continuously expressed spectrum of birth weight, in comparison to values for a 3.2kg birth weight. A similar approach was used, but with logistic regression, to estimate the adjusted odds of having a high percentage (> 85th percentile) of total or truncal fat across the continuously expressed spectrum of birth weight, also when compared to values for a 3.2kg birth weight. All analyses were performed in SAS, version 9.4 (https://www.sas.com), and statistical significance was defined as p < 0.05.

Results
As can be seen in Table 1, 5,490 (55%) participants were women, and 5,510 (55%) declared themselves white, 2,687 (27%) brown, 1,501 (15%) black, 224 (2%) yellow (Asian) and 89 (1%) indigenous. Low birth weight (< 2.5kg) was reported by 529 (5%) and high birth weight (> 4.0kg) by 764 (8%). Median monthly family income per capita was USD 984.79 (25 percentile-75 percentile: 604.65-1,641.32), and 61% of participants had completed college. The majority (54.5%) of mothers of participants, in terms of schooling, had not completed elementary school. A maternal history of diabetes was reported by 1,970 (20%) and a paternal history of diabetes by 1,329 (13%). Mean age was 55.0 ± 8.7 years when bioimpedance analysis was performed. This sample is quite similar to the overall sample of ELSA-Brasil (where 54.4% were women, 52% declared themselves white, 28% brown, 16% black, in second visit mean age was 55.7 ± 9.0, median monthly family income per capita was USD 950.21 (25 percentile-75 percentile: 583.08-1,468.54), 57% of participants had completed college, 57% of mothers of participants had not completed elementary school, low birth weight was reported by 8.4% and high birth weight by 7.5%. Among men, mean BMI was 27.4 ± 4.3kg/m 2 , waist circumference 95.2 ± 11.6cm, hip circumference 100.4 ± 8.0cm, and abdominal height 22.2 ± 3.4cm. Among women, mean BMI was 27.7 ± 5.2kg/m 2 , waist 87.3 ± 12.2cm, hip 103.0 ± 10.3cm, and abdominal height 20.7 ± 4.1cm. Table 1 also presents frequency or mean values of these characteristics by birth weight, separately for men and women. Individuals reporting low birth weight had lower educational attainment and household income per capita. The distribution of maternal and paternal history of diabetes across categories of birth weight was not consistent among men and women. A maternal history of diabetes was greater among those with high birth weight. Figure 1 shows the correlations of our four bioimpedance outcome variables with traditional anthropometric measurements. Absolute measures of bioimpedance (total and truncal fat) correlated strongly with BMI (r ≥ 0.91), abdominal height (r ≥ 0.88), and waist circumference (r ≥ 0.82), with high correlations both in men and women. Relative measures of bioimpedance (total fat/weight, truncal fat/weight) were less correlated, being stronger with BMI (r ≥ 0.77), abdominal height (r ≥ 0.76) and waist/height (r ≥ 0.72). Table 2 describes adjusted means of total body and truncal fat assessed by bioimpedance, according to birth weight and sociodemographic characteristics, separately for men and women. Women had higher values of body fat, both in absolute terms and concerning body weight. Absolute measurements of body fat mass (total and truncal) increased with birth weight (p < 0.001), in both sexes. Fat and truncal fat as percentages of body weight had slightly higher values in the low and high birth weight categories for both men and women, though without statistical significance in comparison with those of appropriate birth weight (p > 0.85). There was no difference regarding sex in these associations (p ≥ 0.87 for the interaction between birth weight and sex).
As also seen in Table 2, older men and women presented greater amounts of both overall and truncal fat, both in relative and absolute terms. However, women who declared being of white/race color had lower mean values of fat indices, while men had higher values. Similar differences by sex were also seen across income and educational attainment; women with higher income and educational attainment having higher values, while men, lower ones. Individuals with a maternal or paternal history of diabetes had uniformly greater indices.
Restricted cubic spline analyses for the 5,212 participants who reported exact birth weights ( Figure 2) show a general U-shaped pattern of mean differences of relative total body and truncal fat across birth weight categories. In these analyses, the weight nadir was always at or near a birth weight of 3.2kg. However, only the greater mean percentage fat values seen with high birth weight achieved statistical significance, as indicated by the low border of the shaded zones of 95% confidence being above the line of zero difference.
To examine the magnitude of the associations of birth weight categories with high (> 85 percentile) values of absolute and relative total and truncal fat, the next logistic regression analyses perfomed were adjusted for study center, gender, age, family income, race/color, participant's and mother's educational attainment. As seen in Figure 3   As also seen in Figure 3 (to the right, for relative amount of fat), although sex interactions were not statistically significant (p > 0.11), women presented U-shaped associations between birth weight and fat as a percentage of body weight, while men showed no association. As seen in the bottom figure, women with low birth weight showed higher odds of presenting a high truncal fat percentage

Discussion
Our measurements of body fat, assessed by bioimpedance in middle-age and elderly subjects, correlated strongly with measurements of BMI, sagittal diameter and waist-to-height ratios. When relating absolute measures of body fat with birth weight, we found generally positive associations. However, when comparing fat as a percentage of body weight, men and women both presented non-linear distributions. Notably, both men and women at birth weight extremes presented proportionally greater fat content, both overall and truncal, as adults, with these associations for high birth weight being statistically significant. However, in parallel analyses considering the relative odds of presenting a high portion of body weight as fat, only women maintained this U-shaped association, showing statistically greater risk of presenting high truncal fat if born with low birth weight, and a trend toward greater Cad. Saúde Pública 2021; 37(2):e00061619 Table 2 Adjusted bioimpedance measures of body fat according to birth weight and socio-demographic characteristics, separately for men and women. Brazilian  risk with high birth weight. To our knowledge, non-linear associations of birth weight with measurements of body composition as determined by bioimpedance have not been previously described. Accumulating evidence regarding the associations of birth weight with overall/central obesity have produced heterogenous results across studies and populations. A systematic review documented a positive association with BMI but an inverse association with truncal obesity assessed by the ratio of Cad. Saúde Pública 2021; 37(2):e00061619

Figure 2
Estimated differences in bioimpedance measures of % body and truncal fat (fat/weight; reference: 3.2kg birth weight), in restricted cubic spline analyses adjusted for study center, sex, age, income, race/color, participant's and mother's educational attainment. Brazilian Longitudinal Study of Adult Health (ELSA-Brasil) (N = 10,011).
Note: black = all (n = 5,212); red = women (n = 2,804); blue = men (n = 2,408). Distributions of birth weight are superimposed. subscapular/triceps skinfolds 14 . Positive associations with adult obesity assessed by body mass index were also documented in a systematic review 9 . The relationship of birth weight with central fatness in adults was further summarized by another systematic review, documenting that, after adjustment for adult BMI, no association was observed for waist circumference, but an inverse one was observed for waist-to-hip ratio 15 . Similar to the findings of that review, we described sex-specific non-linear associations between birth weight and measurements of obesity/central obesity in women when assessed by BMI, waist circumference and wait-to-hip ratio, although only the latter was statistically significant in adjusted models 30 . These findings are consistent with another recent study among Chinese children and adolescents, which, while not presenting sex-specific analyses, also described J-shaped relationships between birth weight and waist-to-height ratios 37    Notes: blue = men; red = women. Analyses were adjusted for study center, sex, age, family income, race/color, participant's and mother's educational attainment.
One should note that most studies describing linear associations between birth weight and anthropometric obesity indices in adults 9 come from populations that are not suffering major fetal and early life undernutrition. Greater risk of chronic diseases associated with low birth weight is evident in settings of fetal and early life undernutrition, such as periods of famine 21,26,38 , highlighting the need to avoid both low and high birth weight. According to Neel's hypothesis, "thrifty genes" selected during Cad. Saúde Pública 2021; 37(2):e00061619 adversities across evolution may predispose to chronic diseases in modern times 39 . An alternative hypothesis, the thrifty phenotype hypothesis, was proposed later, relating intrauterine adversity with future chronic diseases 40,41,42 . Based on studies of subjects born in periods of food restriction and war 21,26,38 , intra-uterine growth retardation was proposed to cause future overall/central obesity. Interestingly, associations were observed among women but not among men 21,26 , which is consistent with sex-specific mechanisms of intrauterine programming for future survival 16,18,20 .
Mechanisms relating intrauterine environment, low and high birth weight, and future obesity/ central obesity include genetic 43 , epigenetic 9 , environmental 44,45 , and nutritional 46,47 factors. However, how these factors interact to explain the paradox of low and high birth weight causing future metabolic risk is still not clear 48 . It seems logical, though, to observe sex-specific patterns of associations such as those pointed out here, with women of low and high birth weight presenting greater proportions of body and truncal fat, as this may have explanations based on evolutionary necessities, with adaptations being present in utero to guarantee survival and reproduction. Throughout the life course, additional factors play a role, notably catch-up growth during childhood 49 and modern life styles which increase susceptibility for obesity, diabetes, and other chronic diseases.
A few potential limitations of our study merit discussion. First, birth weight was obtained by interviews with our adult subjects. Participants initially indicated their estimated birth weight within predefined categories and then provided their best recollection of the exact weight. Although we encouraged them to inquire among relatives and search for documents to confirm the informed birth weight, only 52% were able to provide an exact birth weight. Second, fat mass was obtained by bioimpedance, which assumes constant hydration of the fat-free mass, which may not hold in the presence of disease 50 . However, our standardized protocol of measurement in the morning after an 8-16 hour fast and after voiding exercises minimized this possibility. Additionally, bioimpedance does not measure body fat directly, representing only an estimate, and does not differentiate subcutaneous from visceral fat as many body imaging techniques do 51 .
On the other hand, our study has several strengths. We are unaware of prior studies investigating the relationship of birth weight and adult adiposity that have been able to measure body fat as we, have using bioimpedance. Our expressive sample size and multicentric design carried out in a free-living population expands on previous studies that have explored the subject, which were most conducted with individuals of European ancestry and in a historical and sociocultural situation distinct from that now experienced by most of the world's population, who live in low-and middle-income countries such as Brazil.
In conclusion, we found associations of high birth weight with larger total body and truncal fat mass, in men and women, and U-shaped associations between birth weight and relative total and truncal fat, most notably in women. These results add further and more detailed information to the associations previously described in ELSA-Brasil by anthropometric measures. They suggest that low and high birth weight contribute to adult central obesity, highlighting the need to prevent both when formulating nutritional policies during pregnancy.